CDM-based constitutive model incorporating strength and stiffness degradation for ULCF prediction of weld metal

Existing approaches face challenges in accurately predicting both the mechanical degradation of weld and base metals and their ultra low cycle fatigue life. To address this issue, this study investigates the mechanical degradation behavior of weld metal under varying stress triaxiality conditions through monotonic tensile and large-strain cyclic loading tests. Within a ductile damage mechanics framework, decoupled strength and stiffness correction factors are proposed. Machine learning–assisted regression is then employed to develop evolution equations for the corresponding damage indices, which are subsequently incorporated into a finite element user subroutine to achieve a decoupled modification of the traditional Lemaitre–Chaboche constitutive model. Comparative validation demonstrates that, for the modified model, the average prediction errors for pre-fracture strength and stiffness are reduced to 2.92% and 1.59%, respectively. Furthermore, a regression relationship between the critical damage thresholds and structural parameters is calibrated based on experimental data, and a corresponding ULCF crack initiation criterion is established. Finite element life prediction results yield maximum, mean, and standard errors of 20%, 6.6%, and 0.074, respectively. This study provides a reliable methodological support for damage evolution analysis and life assessment of welded structures subjected to high-strain cyclic loading.